Motion recognition has long been a research topic in the field of ambient assisted living. Radar has a good application value in indoor activity monitoring due to its privacy protection and non-intrusive. In this stud...
详细信息
Motion recognition has long been a research topic in the field of ambient assisted living. Radar has a good application value in indoor activity monitoring due to its privacy protection and non-intrusive. In this study, the authors use an ultra-wideband (UWB) radar with the centre frequency of 7.25 GHz to receive the reflection signal. Time-frequency analysis is performed on the radar signal to attain the time-varying range-Doppler images (TRDI) which represent changes in motion characteristics over time. They then use the principal component analysis (PCA) algorithm or a pre-trained convolutional autoencoder (CAE) to extract features from TRDI. Finally, gated recurrent unit is employed to dynamically model these features and classify different motions. This recognition process can be synchronised with the action flow without having to wait for the motion to complete before starting the classification. The experimental results show that it has an accuracy of 93.06% for the PCA-based method and 96.80% for the CAE-based method in recognising eight kinds of indoor motions, reaching or even exceeding the performance of the non-synchronous algorithms.
暂无评论